I Finally Understand What All the Obstacle Avoidance Hype Is About
ยท fpv_racer_07 · 08/23/2026 11:00 PM
Coming From FPV, This Felt Almost Weird
First thing I noticed is how calm the whole experience is. Like, I'm used to the drone doing exactly what I tell it to do, instantly, no hesitation, no second-guessing. If I nose it into a tree that's on me. The Air 3 though, it's almost like it has opinions. You push it toward something and it just... slows down. Stops. Holds there. The first time it happened I genuinely thought something was wrong with the signal.
The obstacle avoidance on that thing uses APAS 5.0, which is DJI's omnidirectional sensing system. There are sensors on the front, back, sides, top, and bottom. It's using a mix of wide-angle vision sensors and infrared for the bottom. The system is constantly building a map of what's around it and adjusting flight paths in real time. When I flew it through a section of trail with overhanging branches, it threaded through stuff I honestly wasn't confident about. Not perfectly, it did stop once and just hover rather than commit to a path, but still. For a drone that's not being piloted by someone with years of stick time, that's genuinely impressive.
The Transmission Side of Things Is Wild Too
I also spent some time thinking about the O3+ transmission system while I had it. My FPV quads run either analog video or DJI's older digital FPV system, so I have some frame of reference here. The range and signal stability on the Air 3 compared to what I'm used to is on another level. O3+ is doing a ton of work behind the scenes - it's constantly switching between frequencies, adjusting transmission power, basically doing everything it can to maintain a clean link even in congested RF environments.
I flew it about a mile out over an open field just to see what would happen with signal quality. The feed stayed completely clean the entire time. No breakup, no latency spikes. For FPV I'm usually getting some kind of interference or noise by that range depending on where I am. Different use cases obviously, but it made me appreciate how much engineering goes into consumer drones that we kind of just take for granted.
The Battery Situation Is Actually Reasonable Now
One thing I've always kind of rolled my eyes at with consumer drones is the battery specs feeling inflated. Like when a manufacturer says 40 minutes flight time and you get 22 in real world conditions. The Air 3 was getting me pretty consistent 28 to 32 minute flights with active flying, not just hovering. The Intelligent Flight Battery system is smart about discharge and it gives you actual time remaining estimates that feel trustworthy rather than wildly optimistic.
Compare that to my 5-inch freestyle build which runs 4S 1500mAh packs and gets me maybe 4 to 5 minutes of aggressive flying. Totally different world. I understand why consumer drones optimize for efficiency over performance at this point. The battery tech on the higher end consumer stuff has genuinely caught up to where the marketing claims are at least somewhat believable now.
The AI Subject Tracking Stuff Genuinely Surprised Me
I saved this for its own section because it deserves it. I had my dog with me at the trail on Sunday and on a whim I used the subject tracking to lock onto him while he was running around off leash. The drone followed him through some pretty chaotic movement, kept him in frame, adjusted altitude when he went up a small embankment, and didn't lose the lock for a solid three minutes of continuous tracking. I was just standing there holding the controller doing nothing.
The way these tracking algorithms work now is they're not just doing color blob tracking like the early days. The system is doing real-time subject recognition, estimating where the subject is likely to move based on trajectory, and combining that with the obstacle avoidance so it's not flying itself into stuff while following. It's a lot of processing happening on the drone itself, not just offloaded to the controller or an app. The onboard chips have gotten fast enough to handle it.
I'll be honest, as someone who flies FPV it made me feel a little bit like, okay what am I even contributing here. But then I took it back home and strapped on my goggles and flew my quad through a gap the size of a lunch box and felt better about myself. Different tools, different purpose.
Where I Think This Is All Going
This is the part where I'm just kind of thinking out loud because I genuinely don't know. But it seems obvious that the AI and automation side of drone tech is moving faster than the hardware side at this point. The motors and props and frame materials are pretty mature. But the software, the sensors, the onboard processing - that stuff is changing fast.
I've been reading about stuff like neural-network-based depth estimation replacing traditional stereo vision sensors, which could mean better obstacle avoidance in low light or in situations where texture-based sensing falls apart. And there's a lot of work being done on swarm coordination for commercial applications, which is interesting even if it's not relevant to what I do.
For FPV racing specifically I'm curious whether any of this trickles down. There are already some racing drones experimenting with GPS-based return-to-home as a safety net and some are messing with auto-arm sequences. Nothing like what the consumer drones are doing, but the technology exists and it's getting cheaper.
Anyway. This was a longer post than I planned. If you're an FPV pilot who's never touched a consumer drone I'd say borrow one for a weekend if you get the chance. It's a genuinely different experience and it gave me a lot to think about in terms of where the technology is heading. I'll probably write something more specific about FPV transmission systems next since that's more in my wheelhouse. Thanks for reading if you made it this far.
- Read more...
- 0 comments
- 54 views
-